Retry failed and cancelled async predictions instead of caching errors

great_ai/deploy/great_ai.py

@@ -265,7 +265,7 @@ class GreatAI(Generic[T, V]):
                 result = func(*args, **kwargs)
                 return cast(T, t.finalise(output=result))
 
-        @alru_cache(maxsize=get_context().prediction_cache_size)
+        @alru_cache(maxsize=get_context().prediction_cache_size, cache_exceptions=False)
         async def func_in_tracing_context_async(
             *args: Any,
             do_not_persist_traces: bool = False,

great_ai/external/async_lru.py

@@ -148,9 +148,9 @@ def alru_cache(
                     _cache_hit(wrapped, key)
                     return await asyncio.shield(fut)
 
-                exc = fut._exception
+                exc = None if fut.cancelled() else fut.exception()
 
-                if exc is None or cache_exceptions:
+                if not fut.cancelled() and (exc is None or cache_exceptions):
                     _cache_hit(wrapped, key)
                     return fut.result()